Hi Steve,
I'd say it's a "balance" between marketing and any strict numbers vendors may provide. Chipset capacity is wlan PHY capacity. But that's top capacity for an AP that might be e.g. 4x4 11ac 80 MHz, it may even be sum of all radios capacities (ever heard of 11 Gigabit gaming 'wifi router'? :D). First, we go down because of a certain radio being used - so the overall AP capacity fraction is considerable. Then, our client devices are not as capable as an AP. Typically they are 1x1 (smartphones) or 2x2 (laptops). Then, we are not going to use 80 MHz outside home, in something bigger than a small office even 40 MHz might be a no-go. So we end up with 20 MHz channel 1x1 or 2x2 communication. Thus, our PHY datarates are definitely smaller than what AP can handle.
Once we know device's max capabilities, it is a subject to datarate selection (based on some criteria but mainly SNR) - it can be lower than max in certain situation (distance, interferences etc.).
Once we have certain datarate picked (on Windows you can nicely check yours with cmd > netsh wlan show interface and find out it sometimes is different for Tx and Rx), our PHY datarate is subject to a lot more than user traffic. Statistically, I remember a graph saying that something like 25% of frames are data frames. Besides you have beacons, probe requests and responses, other 802.11 management/control frames, every data frame has to be acknowledged upon successful reception... So - depending on PHY version, like 11n or 11ac or 11ax - there is a percentage of PHY datarate we may expect to provide us as a user data throughput. The numbers are different for different PHY but I'd assume 60-75% of datarate is something we may expect with iperf on a single wireless client, IF...
If there's no ACI/CCI interference, if there's only one client/AP on that channel in close proximity trying to speak to the AP/client and so on. The more devices we have connected to the same AP (or more generally - on the same channel, if they can "hear" each other), the more devices contend to use the medium which is a collision domain (like we had in the past with wired networks) and Wi-Fi is half-duplex so devices have to speak in turn. They have pretty complicated CSMA/CA algorithm that takes time and will produce some statistical outcome but I think we can say for an example, 4 devices talking to an AP, all with datarates 400 Mbps, will have the throughput divided into 4 parts if airtime fairness is enabled. In regular deployments something closer to 45% of a datarate is expected to be seen as a throughput, if the design and radio environment works nicely.
More:
https://www.duckware.com/tech/wifi-in-the-us.html#wifispeeds (section 3 and 4)
https://divdyn.com/wi-fi-throughput/
https://mcsindex.net/
https://d2cpnw0u24fjm4.cloudfront.net/wp-content/uploads/802.11n-and-802.11ac-MCS-SNR-and-RSSI.pdf or
https://community.cisco.com/legacyfs/online/attachments/discussion/revolution-wi-fi-mcs-to-snr-singl... Hope that helps,
Tomasz